A large amount of data about bridge conditions and maintenance actions, and related factors, are being collected each year. Such data include bridge inventory data, traffic and weather data, and unstructured textual bridge inspection reports. The wealth of these heterogeneous data from multiple sources offers great promise to data analytics for better predicting bridge deterioration. However, existing data-driven bridge deterioration prediction approaches mostly focus on learning from a single type of data from a single source - mainly the National Bridge Inventory (NBI) data. They are limited in learning from multi-type and multi-source data, which collectively cover a large number of factors that affect the deterioration of bridges. To address this limitation, this paper proposes a novel smart bridge data analytics framework. The framework includes three primary components: (1) information extraction: information about bridge conditions and maintenance actions is extracted from unstructured textual inspection reports; (2) data integration: the data/information extracted from the reports are linked and fused, and integrated with bridge inventory, traffic, and weather data; and (3) data analytics: bridge deterioration is predicted based on the integrated data. This paper focuses on presenting the proposed framework and its preliminary experimental evaluation results. The results show that, by learning from integrated bridge data, the proposed framework achieved an average prediction precision and recall of 82.8% and 78.2%, respectively, compared to 71.5% and 60.2% when only learning from NBI data.